2/21/2018

Data

  • Measurements of relative abundances.
  • Data are collected as counts.
  • Variability in estimate of composition dependent on sample sizes and composition.

Testate amoebae

Testate Data

Pollen Data

Models

  • Weighted Averaging.
  • Modern Analog.
  • Random Forest.
  • BUMMER.
  • Bayesian Methods (MVGP).

BUMMER

  • Univariate response of species to covariate.
  • Gaussian-shaped response curve.

MVGP

  • Flexible response of species to covariate.
  • Shape of response curve determined by data.

MVGP Testate

MVGP Pollen

WA vs. MVGP

Model Prediction Comparison

Caribou Bog Reconstruction

Ecological Functional Groups Testate

Ecological Functional Groups Pollen

Simulation Reconstruction

Poor Analog Problem

Cross-validation Results Testate

Score MVGP GAM WA MAT MLRC
CRPS 0.2959 0.2966 0.2962 0.2876 0.3577
MSPE 0.2793 0.2933 0.2924 0.2497 0.4815
MAE 0.3897 0.3943 0.4016 0.3690 0.4662
95% CI coverage 78.3708 79.7753 94.3820 98.8764 92.4157

Cross-validation Results Pollen

Score MVGP GAM WA MAT MLRC
CRPS 0.2746 0.2828 0.2661 0.2441 0.2877
MSPE 0.2178 0.2119 0.2287 0.1673 0.2722
MAE 0.3864 0.3912 0.3719 0.3121 0.4113
95% CI coverage 87.5000 78.9474 94.0789 100.0000 96.7105

Asymmetric Predictive Distributions